That's from the period when there was no standardization of how the CPU talked to the graphics device. Triangles or quads? Shared memory or command queues? DMA from the CPU side or the graphics device side? Graphics as part of the CPU/memory system or as part of the display system? Can the GPU cause page faults which are serviced by the virtual memory system?
Now we have Vulkan. Vulkan standardizes some things, but has a huge number of options because hardware design decisions are exposed at the Vulkan interface. You can transfer data from CPU to GPU via DMA or via shared memory. Memory can be mapped for bidirectional transfer, or for one-way transfer in either direction. Such transfers are slower than normal memory accesses. You can ask the GPU to read textures from CPU memory because GPU memory is full, which also carries a performance penalty. Or you can be on an "integrated graphics" machine where CPU and GPU share the same memory. Most hardware offers some, but not all, of those options.
This is why a lot of stuff still uses OpenGL, which hides all that.
(I spent a few years writing AutoCAD drivers for devices now best forgotten, and later trying to get 3D graphics to work on PCs in the 1990s. I got to see a lot of graphics boards best forgotten.)
And that was an evolution of earlier 2D cards where you had a potential mixture of CPU-addressable framebuffer and various I/O ports to switch modes between text and raster graphics, adjust video modes in DACs, adjust color palette lookup tables, load fonts for text modes, and maybe address some 2D coprocessors for things like "blitting" (kind of like rectangular 2D DMA), line drawing, or even some basic polygonal rendering with funny options like dithering or stipple shading...
And that is because of that low level exposure, and extension spaghetti, that Khronos finally acknowledged at Vulkanised 2025 that adoption didn't went on as expected, and there is now a roadmap being put into place on how to improve the whole situation.
Just looked at that roadmap.[1] Good to know they are thinking about that.
I've discussed this in the context of Rust renderers. There are about four of them, and they all sit on top of Vulkan and export roughly similar APIs.
The API is roughly comparable to three.js - meshes, textures, materials, lights, shadows, and updating are supported.
They all get to the point where they can load static glTF scenes, and then the devs hit the hard concurrency problems and bail.
That's a much more comfortable level for users who need to get something done in 3D. Vulkan level belongs to people who write engines.
Getting off topic, though. This is not a retro problem. This is a current problem.
Further discussion on the Vulkan roadmap request for comments on Discord, here.[2]
Did you ever play around with the Matrox Digisuite SDK for DirectX for real time video processing, that had a visual data flow programming language interface for its video processing nodes that would execute inside the gpu?
>1996: Matrox launches the Matrox Genesis Series vision processor, and with its highly pipelined, parallel-processing architecture, establishes a new performance industry standard.
>The company also introduces Matrox DigiSuite, the first affordable all-digital system for real-time, non-linear editing. The award-winning DigiSuite breaks new ground in offering a full set of professional-quality editing and effects capabilities. DigiSuite becomes the world's best-selling professional video editing card.
This kind of retrospective from key people who were involved is invaluable from an historical perspective. I find hearing first-hand accounts of the context, assumptions, thought processes, internal debates, technical limitations, business realities and even dumb luck a good way to not only understand how we got here but how to do as well (or better) going forward.
While the nitty gritty detail of recollections captured when still fresh in memory can be fascinating, I especially appreciate reflections written a few decades later as it allows putting the outcomes of key decisions in perspective, as well as generally enabling more frank assessments thanks to fewer business and personal concerns.
I'm excited about this too, but it's a little concerning there's a brand in the title. There's no shortage of those from ati, intel, amd, apple, ibm, the game gaggle, etc to interview. The fact that nvidia succeeded where others failed is largely an artifact of luck.
I would say nVidia made their own luck. A lot of their success can be attributed to their management never losing sight of the fact that the software is just as important as the hardware. Both drivers and CUDA are key to nVidia's success. ATI and nVidia would trade places on quality of hardware, but there was never a question on the software side.
I'm not sure how nvidia's driver track record would have helped them, but drivers nor linux nor software in any way has ever really been nvidia's strong-suit. but even with the popularity of it CUDA cannot explain nvidia's success alone; you also need the demand of butcoin and the secondary-but-farcical imitation of LLMs but also the inexplicable lack of awareness of alternatives that need explaining...
I worked on CUDA and OpenCL in the 2010-2014 timeframe, well before buttcoin and LLMs were profit centers, and Nvidia was already well ahead in the "GPUs as general compute" area. Literally everyone doing highly parallel HPC wanted to use Nvidia, despite AMD having higher throughout for some workloads. It was better, easier to use software.
I'll add to that: even though it is true that "drivers nor linux nor software in any way has ever really been nvidia's strong-suit", as GP put it, their software was still miles ahead of its competitors. In the land of the blind a one-eyed man is king, and all that.
I disagree with "largely". Luck is always a factor in business success and there are certainly some notable examples where luck was, arguably, a big enough factor that "largely" would apply - like Broadcast.com's sale to Yahoo right at the peak of the .com bubble. However, I'm not aware of evidence luck was any more of a factor in NVidia's success than the ambient environmental constant it always is for every business. Luck is like the wind in competitive sailing - it impacts everyone, sometimes positively, sometimes negatively.
Achieving and then sustaining substantial success over the long run requires making a lot of choices correctly as well as top notch execution. The key is doing all of that so consistently and repeatedly that you survive long enough for the good and bad luck to cancel each other out. NVidia now has over 30 years of history through multiple industry-wide booms, downturns and fundamental technology transitions - a consistent track record of substantial, sustained success so long that good luck can't plausibly be a significant factor.
That said, to me, this article didn't try to explain NVidia's long-term business success. It focused on a few key architectural decisions made early on which were, arguably, quite risky in that they could have wasted a lot of development on capabilities which didn't end up mattering. However, they did end up paying off and, to me, the valuable insight was that key team members came from a different background than their competitors and their experiences with multi-user, multi-tasking, virtualized mini and mainframe architectures caused them to believe desktop architectures would evolve in that direction sooner rather than later. The takeaway being akin to "skate to where the puck is going, not where it is." In rapidly evolving tech environments, making such predictions is greatly improved when the team has both breadth and depth of experience in relevant domains.
I remember sitting next to David Rosenthal at a conference reception (must have been FAST, which makes sense given his involvement with LOCKSS) in San Jose some time around 2010 or 2011, not knowing up front who he was. He explained some of the innovations he had made at NVIDIA around making the hardware more modular and easier for parallel teams to work on, and we chatted about the rumors I had heard about SUN thinking about licensing the Amiga hardware, which he confirmed but said would have been a bad idea, because the hardware didn't support address space protection. I guess I didn't know enough about him or NVIDIA to be sufficiently impressed at the time, but he was a very friendly and down to earth person.
That's right, he's a great down-to-earth guy (but he can still write like a passionate punk rocker -- see below), and there's a wealth of interesting thoughtful stuff on his blog. I've known him since the days of the X10 / X11 / NeWS window system wars.
He worked with James Gosling on Andrew at CMU and NeWS at Sun, and on X10 as well as X11 and ICCCM, and he implemented the original X10 compatibility layer that was in NeWS 1.0, before X11 was a "thing".
One of my favorite classics is his Recreational Bugs talk [1989] by "Sgt." David Rosenthal (author of the ICCCM, developer of the Andrew Window Manager, X10, X11, and NeWS, employee #4 and chief scientist at Nvidia):
At the 1985 Alvey Workshop "Methodology of Window Management", David Rosenthal and James Gosling presented "System Aspects of Low-Cost Bitmapped Displays" and "A Window Manager for Bitmapped Displays and Unix" about their work on Andrew at CMU. And James presented "SunDew - A Distributed and Extensible Window System" about his work at Sun, which was later renamed NeWS, and was what convinced David to leave CMU and join him at Sun.
Warren Teitelman's "Ten Years of Window Systems - A Retrospective View" covers the fascinating history of Smalltalk, DLisp (Interlisp), Interlisp-D, Tajo (Mesa Development Environment), Docs (Cedar), Viewers (Cedar), SunWindows and SunDew systems:
Here's David Rosenthal's notorious Sun Deskset Environment flame that some rogue leaked to the Unix-Haters mailing list (inspiring the Unix-Haters Handbook's X-Windows chapter), in which he poignantly concluded:
"It's like having a Roy Lichtenstein painting on your bedroom wall.":
From: DR
Date: 18 Oct 90 17:02:39 GMT
Newsgroups: sun.open-windows
Subject: Re: Deskset environment
[NS replied to me directly. Her reply illustrates the
reasons why I sent out yesterday's mail so perfectly that
I'm taking the liberty of copying my reply to
openwindows-interest]
> When we give standard Deskset presentations, a couple of
> things tend to "dazzle" the audience ...
>
> 1. Use the MT Calendar template to generate an
> appointment. Mail it to yourself, then
> drop it onto CM which will schedule it. The
> template is totally hokey (we're working on
> it) but it works and is wizzy.
>
> 2. Build a small application with GUIDE and make it
> on the spot. Show it up and running on XView
> in minutes. You can talk to BW about that
>
Thank you, but you have completely missed the point. I
don't want to show people how whizzy the standard default
desktop environment is. That's your job.
I want to give a talk about a quite different subject. I
merely want to *use* the desktop environment to achieve my
own ends. And as soon as I try to actually *use* it for
something instead of merely showing off the glitz, it falls
to pieces in my hands. Unfortunately, this is becoming all
too common in Sun products these days, because we no longer
*use* the things we build for anything but whizzy demos.
Have you ever actually tried to *use* the desktop for
anything? Like, say, printing a PostScript file? The
answer has to be no - because dropping a PostScript file on
the print tool doesn't work. Or binding a shell command to
a pattern? Again no, because doing so depends on
undocumented features of /etc/filetype. Even trying to
create a new icon from the standard set causes the icon
editor to dump core. I'm not joking when I say that I've
been filing a bug report every couple of hours of trying to
use the desktop. Its this kind of fragility that shows me
that I'm treading on fresh snow. No-one else has walked
this way.
And that is a truly sad commentary on the state of Sun -
no-one has been this way because no-one believes that
there's anything worth doing over this way. The reason Unix
was such an advance over previous operating systems was that
you could customize your environment in arbitrary ways.
With just a few shell scripts, for example. Its just like
the cold war - in our anxiety to compete with the enemy
we've ended up losing the things that made our way of life
worth defending in the first place. Like the freedom to
disagree with the authorities.
> I believe you're correct in saying that most people live
> with the default environment, but I think it's only partly
> because they don't know how to customize it. We've done
> some user testing and, surprisingly, people either prefer
> the default environment or just don't want to take the
> time to make it special. This is particularly true of
> people like admins, marketing, etc.
Testing whether people actually do customize their
environment is beside the point. Of course they don't. In
order to do it, I have to write C code using bizarre
features of Xview, exercise all my shell wizardry, and
dredge up undocumented features of the system from the
source. And you're suprised when admins can't do this? I
don't expect admins to do it. But I do expect ISVs and
Sun's SEs to be able to do it, and right now they can't.
PS - I notice that someone filed a bug today pointing out
that even your example of dropping a mail message on CM
doesn't work if CM is closed. That's a symptom of the kind
of arrogance that all the deskset tools seem to show -
they're so whizzy and important that they deserve acres of
screen real estate. Why can't they just shut up and do
their job efficiently and inconspicuously? Why do they have
to shove their bells and whistles in my face all the time?
They're like 50's American cars - huge and covered with
fins. What I want is more like a BMW, small, efficient,
elegant and understated. Your focus on the whizzy demos may
look great at trade shows, but who wants to have their tools
screaming at them for attention all the time? It's like
having a Roy Lichtenstein painting on your bedroom wall.
> all an application could do was to invoke methods on virtual objects .. the application could not know whether the object was implemented in hardware or in the resource manager's software. The flexibility to make this decision at any time was a huge advantage. As Kim quotes Michael Hara as saying:
> “This was the most brilliant thing on the planet. It was our secret sauce. If we missed a feature or a feature was broken, we could put it in the resource manager and it would work.”
Absolutely brilliant. Understand the strengths and weaknesses of your tech (slow/updateable software vs fast/frozen hardware) then design the product so a missed deadline won’t sink the company. A perfect combo of technically savvy management and clever engineering.
Interesting that the bus bandwidth was such a bottleneck, when you could also argue it was VRAM. I never looked into why "Hardware transform&lighting" was introduced so late (I'm guessing it didn't fit on the die next to the rasterization stuff?), but I think hardware t&l combined with storing geometry in VRAM would have reduced bus usage to a minimum.
I had a Riva TNT at the time, and a friend's Voodoo... 2? sure ran Half-Life better than my card. The Voodoo 2 probably really was the better GPU at the time. Later games required later APIs (unsupported by older cards) - and much higher performance - anyway, so longevity was not much either way.
This reads as one of the many engineering marvel stories (e.g. Bell Labs, Xerox) where revolutionary technology is created by a combination of (a) clever engineers with enough "free" time, and (b) no clueless managers around.
You can read in Kernighans History of Unix that really good managers - "enlightened management" IIRC - were involved and not just involved, some of them were absolutely crucial or Unix won't have existed. It's not like you can just let loose a couple of big brains and things will work out fine. They won't (and didn't).
Really fascinating story—thanks for sharing! Graphics programming has been a major driving force behind the widespread adoption of object-oriented programming, and the abstraction of devices in this context is truly elegant.
Bruce Factor and Curtis Priem developed a flight simulator called Aviator for Sun's S-Bus GX graphics accelerator. I had one of them on my SS2, and owned a copy of Aviator on CDROM, and loved to play it.
It was one of the earliest Sun games distributed with a free playable demo that you could unlock by buying a license key. It would let you play for a minute or so and then the simulation would stop abruptly. That inspired me to do the same thing with SimCity for the Sun: you could play for a few minutes, then the simulator would switch to a cellular automata that would scramble your city, until you bought a license key to unlock it.
The GX was an 8-bit display, so it uses a 256 color palette, and drew shaded colors by halftone stippling. You can clearly see that in this video:
Aviator 1.5.1 running on SunOS 4.1.4 on a Sparcstation 20
Aviator had "aircraftcap" file in the format of (yech) a termcap file, that allowed you to configure your aircraft, and several fighter aircraft, small planes, and a 747 were included.
I heard a story about a fighter pilot who played Aviator, but he noticed something was wonky about how a plane he was familiar with flew. So he too a look at the aircraftcap file, and was able to find the problem, tweak it, and then it flew much more to his satisfaction!
You could dogfight over the network, and the sneaky thing to do was to hack your aircraftcap file to mount your guided missiles far behind you fighter jet, so when somebody was on your tail, you could shoot them in the ass!
Software in silicon: the methodology behind Sun's GX graphics accelerator architecture:
Reddit: A request to owners of Sun GX hardware such as SPARCstations: Can anyone post screenshots or footage from the late 80s/early 90s flight sim "Aviator"?
The sentence and paragraph which makes it clear that this was megabytes and not kilobytes
At a time when PC memory maxed out at 640 megabytes, the fact that the PCI bus could address 4 gigabytes meant that quite a few of its address bits were surplus. So we decided to increase the amount of data shipped in each bus cycle by using some of them as data. IIRC NV1 used 23 address bits, occupying 1/512th of the total space. 7 of the 23 selected one of the 128 virtual FIFOs, allowing 128 different processes to share access to the hardware. We figured 128 processes was plenty.
Okay but "640" is a completely fictitious number for installed RAM in any given PC.
PC memory was nearly always sold in powers of two. So you could have SIMMs in capacity of 1MiB, 2MiB, 4, 8, 16MiB. You could usually mix-and-match these memory modules, and some PCs had 2 slots, some had 4, some had a different number of slots.
So if you think about 4 slots that can hold some sort of maximum, we're thinking 64MiB is a very common maximum for a consumer PC, and that may be 2x32 or 4x16MiB. Lots of people ran up against that limit for sure.
640MiB is an absurd number if you think mathematically. How do you divide that up? If 4 SIMMs are installed, then their capacity is 160MiB each? No such hardware ever existed. IIRC, individual SIMMs were commonly maxed at 64MiB, and it was not physically possible to make a "monster memory module" larger than that.
Furthermore, while 64MiB requires 26 bits to address, 640MiB requires 30 address bits on the bus. If a hypothetical PC had 640MiB in use by the OS, then only 2 pins would be unused on the address bus! That is clearly at odds with their narrative that they were able to "borrow" several more!
This is clearly a typo and I would infer that the author meant to write "64 megabytes" and tacked on an extra zero, out of habit or hyperbole.
The article is a touch confusing, but I'm pretty sure I agree that they meant the 640 kilobyte limit of the OG PC architecture. The Pentium II dates from 1997, the NV1 to 1995, and the new PCI bus with its whopping 32-bits to 1992. 640MiB would have been a prodigious amount of memory at the time of launch.
I don't think any mathematical relationship between the address bus and either 640KiB or 640MiB was intended, it was just the anchor point for how huge 4GiB of addressing was viewed at the time.
The article then goes on to say that the NV1 used 23 bits of the address bus but adds in the next paragraph that 16 bits remained to use for data. That math isn't working out for me.
Actually, I'm really struggling to understand how this scheme would work at all. It strongly implies open addressing with no other MMIO devices to conflict with, but that's just not how I thought PCI worked. Maybe someone who knows more can explain it to me.
My reading was 640MiB was seen as some extraordinary upper bound that was unlikely to be breached in 1995 leaving lot of bits of the address bus for NV1 for quite some time. The 640 KiB seems definitely not a limit as even IBM PC/AT released in 1984! had an upper limit of 16MiB. So, as NV1's designer you could not assume 640KiB was some sort of upper bound on PC's of 1995 when designing the scheme. As to why 640MiB and not something else, I believe Windows 95 could address 2GB in theory but would start becoming unstable around 512MiB so maybe he chose 640MiB.
The whole thing a bit ironic since Bill Gates took great pains to say he never said 640KiB is all you need (or something like that). Given my example of the IBM PC/AT it definitely was not common understanding of upper limits in 1995 apocryphal or not.
Yeah, I dunno. Besides being a lot for 1995, the address space stuff, if taken at face value, means you'd only have to knick 3 bits off before starting to eat into it. Shrug
By 1995, PCI had vanquished VL-Bus and a cheap entry PC would have had a PCI bus and a 486 CPU at 66 MHz or 100 MHz with 8 MB of RAM, while a decent PC would have had a Pentium with 16 MB of RAM.
The days of 640 kB memory were long forgotten.
Even in 1990, a decent PC would have had a 386 CPU with 4 MB of RAM, which would have been used as extended memory by the MS-DOS programs, while the cheapest computers would have been PC/XT clones with a NEC V20 CPU and with 2 MB of expanded memory, so still well beyond the 640 kB.
It did, especially as the drives got older and shrieky. Plus the floppy and optical drives spinning up. But... funny story, when I moved to Silicon Valley, SSDs didn’t exist yet, but compile times were IO-bound. So my company bought everyone Dell workstations with two 15k RPM SAS drives in RAID 0. When I’d kick off a build, the desk would start vibrating under my fingers.
It made everything feel real!
Later, the company switched new employees to 3x 10,000 RPM SATA drives. Not quite as grindy, but still loud.
Looks like ‘97. Unfortunately I can’t find the receipts! If I had to guess, I think I’d say somewhere around $2000 in all. That computer lasted me a LONG time. When it was done being a desktop, it became a Linux server until I sold it around 2009. Sometimes I wish I’d kept it: the Pentium 2 was such an important CPU and I smile every time I see the one on display in the Computer History Museum.
Alright then! Humbly, I stand corrected about my poor speculation without research. It looks like 640MiB has been a perfectly achievable configuration, especially with 2x256+2x64 or such. That is, I must say, a huge amount of RAM. Like, way more than any video game ever specified in HW requirements. What use cases applied that you could use up 640MiB in that era, I wouldn't know!
I remain a bit mystified about why it would be a hard maximum, though. Did such motherboards prevent the user from installing 4x256MiB for a cool 1GiB of DRAM? Was the OS having trouble addressing or utilizing it all? 640MiB is not a mathematical sort of maximum I was familiar with from the late 1990s. 4GiB is obviously your upper limit, with a 32-bit address bus... and again, if 640MiB were installed, that's only 2 free bits on that bus.
So I'm still a little curious about this number being dropped in the article. More info would be enlightening! And thank you for speaking up to correct me! No wonder it was down-voted!
I did a bunch of media and software development back then so RAM helped a lot. Why 640? Not sure. My particular board could have gone up to 768. I did some googling and found some boards that maxed out at 1 GB.
That was a weird time in computing. Things were getting fast and big quickly (not that many years later, I built a dual-socket Xeon at 2.8 GHz, and before that my brother had a dual socket P3 at 700 MHz.) but all the expansion boards were so special-purpose. I remember going out of my way to pick a board with something like seven expansion slots.
But I think your question about why the author said 640 is fair! Maybe they had a machine like mine around then. Or maybe it’s something NVIDIA was designing around?
As another poster has said, around 2000/2001 there were many PC motherboards for which the maximum RAM capacity was 640 MB of SDRAM (though depending on the motherboard the capacity limit could have other values, e.g. 512 MB or 768 MB were other frequent values). Only around 2003, motherboards where you could install up to 4 GB of DDR memory became widespread.
I had a few such motherboards, with Pentium III "Coppermine" and with AMD Athlon "Thunderbird" (with 3 DIMM sockets, 2 accepting 256 MB modules and 1 accepting 128 MB modules).
While I had initially equipped them with less memory, i.e. either 128 MB or 256 MB, because of the high cost at that time, eventually I had maxed up the installed memory to 640 MB on the Athlon motherboards.
At that time, the memory controller was not included in the CPU, but it was contained in an external chip, the Northbridge. The MB chipsets could come from various vendors, including Intel and Via. Depending on the chipset, different maximum amounts of SDRAM could be supported. Sometimes there were also limits for the maximum amount of RAM caused by the L2 cache controller, due to the size of the tags required in cache lines. Even with the same chipset, the MB manufacturer could choose to route different numbers of address lines to the sockets, probably depending on whether the extra capacitive loading caused by a bigger memory module would have allowed it to function at the rated speed on that PCB.
64 MB would have been a typical maximum memory around 1998 (2 sockets for 32 MB SDRAM modules, on a MB with an AGP socket for the graphics card), so if the author would have had that year in mind, you would be right.
For a while, I had 36MB on a 486 mobo, with a combination of SIMMs and DIMMs. As others have mentioned, 640MB is a valid, if odd, configuration for RAM.
QUESTION: I read in a newspaper that in 1981 you said, ``640K of memory should
be enough for anybody.'' What did you mean when you said this?
ANSWER: I've said some stupid things and some wrong things, but not that. No
one involved in computers would ever say that a certain amount of memory is
enough for all time.
The need for memory increases as computers get more potent and software gets
more powerful. In fact, every couple of years the amount of memory address
space needed to run whatever software is mainstream at the time just about
doubles. This is well-known.
When IBM introduced its PC in 1981, many people attacked Microsoft for its
role. These critics said that 8-bit computers, which had 64K of address space,
would last forever. They said we were wastefully throwing out great 8-bit
programming by moving the world toward 16-bit computers.
We at Microsoft disagreed. We knew that even 16-bit computers, which had 640K
of available address space, would be adequate for only four or five years. (The
IBM PC had 1 megabyte of logical address space. But 384K of this was assigned
to special purposes, leaving 640K of memory available. That's where the
now-infamous ``640K barrier'' came from.)
A few years later, Microsoft was a big fan of Intel's 386 microprocessor chip,
which gave computers a 32-bit address space.
Modern operating systems can now take advantage of that seemingly vast
potential memory. But even 32 bits of address space won't prove adequate as
time goes on.
Meanwhile, I keep bumping into that silly quotation attributed to me that says
640K of memory is enough. There's never a citation; the quotation just floats
like a rumor, repeated again and again.
It's hard to tell exactly what time frame the author is referencing there. For context, NV1 was released in '95, by which time it was not uncommon for a new PC to have 8-16 MB of memory (I had a 486 with 16 MB by '94). Especially if you planned to use it for gaming.
Maybe that is what they were thinking but anything designed to work with a PCI bus would have been introduced after PCs became capable of using more memory than that.
> Because Nvidia became one of the most valuable companies in the world, there are now two books explaining its rise and extolling the genius of Jensen Huang,
Yeah, he's a real genius. (Sarcasm). He is a marking guy, there is no genius behind this man.
The fact that Nvidia uses its market position cause harm to the industry by strong-arming partners to toe the line makes this company a problem, just like all the others. They operate like any other predatory corporation.
You would be wrong. He worked at AMD as a design engineer, and later went to LSI logic helping customers put out custom ASICs. One of his customers, a big customer, was Sun, helping with their SPARC processor and the GX graphics chips, and no doubt many others.
In 1989-1991 I did three ASICs at LSI Logic -- and Jensen was my liason there on the latter two. He was incredibly smart, hard working, technically knowledgeable, kind, patient, and generous with his time despite being very busy.
The marketing stuff came later (or maybe said better: it was latent and it came out later)
Probably read the rest? I did not see Jensen's name on any of the patents that this key engineer discusses the detail and rationale of, and I feel that those names are listed fairly deliberately.
That's from the period when there was no standardization of how the CPU talked to the graphics device. Triangles or quads? Shared memory or command queues? DMA from the CPU side or the graphics device side? Graphics as part of the CPU/memory system or as part of the display system? Can the GPU cause page faults which are serviced by the virtual memory system?
Now we have Vulkan. Vulkan standardizes some things, but has a huge number of options because hardware design decisions are exposed at the Vulkan interface. You can transfer data from CPU to GPU via DMA or via shared memory. Memory can be mapped for bidirectional transfer, or for one-way transfer in either direction. Such transfers are slower than normal memory accesses. You can ask the GPU to read textures from CPU memory because GPU memory is full, which also carries a performance penalty. Or you can be on an "integrated graphics" machine where CPU and GPU share the same memory. Most hardware offers some, but not all, of those options.
This is why a lot of stuff still uses OpenGL, which hides all that.
(I spent a few years writing AutoCAD drivers for devices now best forgotten, and later trying to get 3D graphics to work on PCs in the 1990s. I got to see a lot of graphics boards best forgotten.)
And that was an evolution of earlier 2D cards where you had a potential mixture of CPU-addressable framebuffer and various I/O ports to switch modes between text and raster graphics, adjust video modes in DACs, adjust color palette lookup tables, load fonts for text modes, and maybe address some 2D coprocessors for things like "blitting" (kind of like rectangular 2D DMA), line drawing, or even some basic polygonal rendering with funny options like dithering or stipple shading...
It all started with the 8514/A... and maybe a bit of the VGA too.
And that is because of that low level exposure, and extension spaghetti, that Khronos finally acknowledged at Vulkanised 2025 that adoption didn't went on as expected, and there is now a roadmap being put into place on how to improve the whole situation.
The talk being "The Road to The Future".
Just looked at that roadmap.[1] Good to know they are thinking about that.
I've discussed this in the context of Rust renderers. There are about four of them, and they all sit on top of Vulkan and export roughly similar APIs. The API is roughly comparable to three.js - meshes, textures, materials, lights, shadows, and updating are supported. They all get to the point where they can load static glTF scenes, and then the devs hit the hard concurrency problems and bail.
That's a much more comfortable level for users who need to get something done in 3D. Vulkan level belongs to people who write engines.
Getting off topic, though. This is not a retro problem. This is a current problem.
Further discussion on the Vulkan roadmap request for comments on Discord, here.[2]
[1] https://vulkan.org/user/pages/09.events/vulkanised-2025/T2-T...
[2] https://discord.com/channels/427551838099996672/115155677637...
Did you ever play around with the Matrox Digisuite SDK for DirectX for real time video processing, that had a visual data flow programming language interface for its video processing nodes that would execute inside the gpu?
https://video.matrox.com/en/about/history-of-innovation
>1996: Matrox launches the Matrox Genesis Series vision processor, and with its highly pipelined, parallel-processing architecture, establishes a new performance industry standard.
>The company also introduces Matrox DigiSuite, the first affordable all-digital system for real-time, non-linear editing. The award-winning DigiSuite breaks new ground in offering a full set of professional-quality editing and effects capabilities. DigiSuite becomes the world's best-selling professional video editing card.
This kind of retrospective from key people who were involved is invaluable from an historical perspective. I find hearing first-hand accounts of the context, assumptions, thought processes, internal debates, technical limitations, business realities and even dumb luck a good way to not only understand how we got here but how to do as well (or better) going forward.
While the nitty gritty detail of recollections captured when still fresh in memory can be fascinating, I especially appreciate reflections written a few decades later as it allows putting the outcomes of key decisions in perspective, as well as generally enabling more frank assessments thanks to fewer business and personal concerns.
I'm excited about this too, but it's a little concerning there's a brand in the title. There's no shortage of those from ati, intel, amd, apple, ibm, the game gaggle, etc to interview. The fact that nvidia succeeded where others failed is largely an artifact of luck.
I would say nVidia made their own luck. A lot of their success can be attributed to their management never losing sight of the fact that the software is just as important as the hardware. Both drivers and CUDA are key to nVidia's success. ATI and nVidia would trade places on quality of hardware, but there was never a question on the software side.
I'm not sure how nvidia's driver track record would have helped them, but drivers nor linux nor software in any way has ever really been nvidia's strong-suit. but even with the popularity of it CUDA cannot explain nvidia's success alone; you also need the demand of butcoin and the secondary-but-farcical imitation of LLMs but also the inexplicable lack of awareness of alternatives that need explaining...
I worked on CUDA and OpenCL in the 2010-2014 timeframe, well before buttcoin and LLMs were profit centers, and Nvidia was already well ahead in the "GPUs as general compute" area. Literally everyone doing highly parallel HPC wanted to use Nvidia, despite AMD having higher throughout for some workloads. It was better, easier to use software.
I'll add to that: even though it is true that "drivers nor linux nor software in any way has ever really been nvidia's strong-suit", as GP put it, their software was still miles ahead of its competitors. In the land of the blind a one-eyed man is king, and all that.
The author worked at Nvidia and explains some of the engineering decisions made at the time. Why shouldn’t the brand be in the title?
> largely an artifact of luck.
I disagree with "largely". Luck is always a factor in business success and there are certainly some notable examples where luck was, arguably, a big enough factor that "largely" would apply - like Broadcast.com's sale to Yahoo right at the peak of the .com bubble. However, I'm not aware of evidence luck was any more of a factor in NVidia's success than the ambient environmental constant it always is for every business. Luck is like the wind in competitive sailing - it impacts everyone, sometimes positively, sometimes negatively.
Achieving and then sustaining substantial success over the long run requires making a lot of choices correctly as well as top notch execution. The key is doing all of that so consistently and repeatedly that you survive long enough for the good and bad luck to cancel each other out. NVidia now has over 30 years of history through multiple industry-wide booms, downturns and fundamental technology transitions - a consistent track record of substantial, sustained success so long that good luck can't plausibly be a significant factor.
That said, to me, this article didn't try to explain NVidia's long-term business success. It focused on a few key architectural decisions made early on which were, arguably, quite risky in that they could have wasted a lot of development on capabilities which didn't end up mattering. However, they did end up paying off and, to me, the valuable insight was that key team members came from a different background than their competitors and their experiences with multi-user, multi-tasking, virtualized mini and mainframe architectures caused them to believe desktop architectures would evolve in that direction sooner rather than later. The takeaway being akin to "skate to where the puck is going, not where it is." In rapidly evolving tech environments, making such predictions is greatly improved when the team has both breadth and depth of experience in relevant domains.
Nvidia’s Cg language made developers prefer their hardware, I’d say.
Which influenced HLSL, with their close collaboration with Microsoft on DirectX.
[dead]
[dead]
I remember sitting next to David Rosenthal at a conference reception (must have been FAST, which makes sense given his involvement with LOCKSS) in San Jose some time around 2010 or 2011, not knowing up front who he was. He explained some of the innovations he had made at NVIDIA around making the hardware more modular and easier for parallel teams to work on, and we chatted about the rumors I had heard about SUN thinking about licensing the Amiga hardware, which he confirmed but said would have been a bad idea, because the hardware didn't support address space protection. I guess I didn't know enough about him or NVIDIA to be sufficiently impressed at the time, but he was a very friendly and down to earth person.
That's right, he's a great down-to-earth guy (but he can still write like a passionate punk rocker -- see below), and there's a wealth of interesting thoughtful stuff on his blog. I've known him since the days of the X10 / X11 / NeWS window system wars.
He worked with James Gosling on Andrew at CMU and NeWS at Sun, and on X10 as well as X11 and ICCCM, and he implemented the original X10 compatibility layer that was in NeWS 1.0, before X11 was a "thing".
https://en.wikipedia.org/wiki/David_S._H._Rosenthal
One of my favorite classics is his Recreational Bugs talk [1989] by "Sgt." David Rosenthal (author of the ICCCM, developer of the Andrew Window Manager, X10, X11, and NeWS, employee #4 and chief scientist at Nvidia):
https://blog.dshr.org/2018/05/recreational-bugs.html
>"You will get a better Gorilla effect if you use as big a piece of paper as possible." -Kunihiko Kasahara, Creative Origami.
Here's his review of "The Philosopher of Palo Alto", a book about Mark Weiser:
https://blog.dshr.org/2023/06/the-philosopher-of-palo-alto.h...
At the 1985 Alvey Workshop "Methodology of Window Management", David Rosenthal and James Gosling presented "System Aspects of Low-Cost Bitmapped Displays" and "A Window Manager for Bitmapped Displays and Unix" about their work on Andrew at CMU. And James presented "SunDew - A Distributed and Extensible Window System" about his work at Sun, which was later renamed NeWS, and was what convinced David to leave CMU and join him at Sun.
System Aspects of Low-Cost Bitmapped Displays:
https://www.chilton-computing.org.uk/inf/literature/books/wm...
A Window Manager for Bitmapped Displays and Unix (with the first known use of the term "Gorilla effect" as it applies to computer graphics):
https://www.chilton-computing.org.uk/inf/literature/books/wm...
SunDew - A Distributed and Extensible Window System:
https://www.chilton-computing.org.uk/inf/literature/books/wm...
Warren Teitelman's "Ten Years of Window Systems - A Retrospective View" covers the fascinating history of Smalltalk, DLisp (Interlisp), Interlisp-D, Tajo (Mesa Development Environment), Docs (Cedar), Viewers (Cedar), SunWindows and SunDew systems:
https://www.chilton-computing.org.uk/inf/literature/books/wm...
Here's David Rosenthal's notorious Sun Deskset Environment flame that some rogue leaked to the Unix-Haters mailing list (inspiring the Unix-Haters Handbook's X-Windows chapter), in which he poignantly concluded:
"It's like having a Roy Lichtenstein painting on your bedroom wall.":
> all an application could do was to invoke methods on virtual objects .. the application could not know whether the object was implemented in hardware or in the resource manager's software. The flexibility to make this decision at any time was a huge advantage. As Kim quotes Michael Hara as saying:
> “This was the most brilliant thing on the planet. It was our secret sauce. If we missed a feature or a feature was broken, we could put it in the resource manager and it would work.”
Absolutely brilliant. Understand the strengths and weaknesses of your tech (slow/updateable software vs fast/frozen hardware) then design the product so a missed deadline won’t sink the company. A perfect combo of technically savvy management and clever engineering.
Interesting that the bus bandwidth was such a bottleneck, when you could also argue it was VRAM. I never looked into why "Hardware transform&lighting" was introduced so late (I'm guessing it didn't fit on the die next to the rasterization stuff?), but I think hardware t&l combined with storing geometry in VRAM would have reduced bus usage to a minimum.
I wanted to buy a Voodoo card, and due to PCI incompatible version, had to trade it back for a Riva TNT.
Back then I was quite p***d not being able to keep the Voodoo, how little did I know how it was going to turn out.
We hold you singularly responsible for the eventual failure of the Voodoo3/4/5 and Nvidia domination.
Sorry.... :)
I had a Riva TNT at the time, and a friend's Voodoo... 2? sure ran Half-Life better than my card. The Voodoo 2 probably really was the better GPU at the time. Later games required later APIs (unsupported by older cards) - and much higher performance - anyway, so longevity was not much either way.
This reads as one of the many engineering marvel stories (e.g. Bell Labs, Xerox) where revolutionary technology is created by a combination of (a) clever engineers with enough "free" time, and (b) no clueless managers around.
You can read in Kernighans History of Unix that really good managers - "enlightened management" IIRC - were involved and not just involved, some of them were absolutely crucial or Unix won't have existed. It's not like you can just let loose a couple of big brains and things will work out fine. They won't (and didn't).
Really fascinating story—thanks for sharing! Graphics programming has been a major driving force behind the widespread adoption of object-oriented programming, and the abstraction of devices in this context is truly elegant.
Bruce Factor and Curtis Priem developed a flight simulator called Aviator for Sun's S-Bus GX graphics accelerator. I had one of them on my SS2, and owned a copy of Aviator on CDROM, and loved to play it.
It was one of the earliest Sun games distributed with a free playable demo that you could unlock by buying a license key. It would let you play for a minute or so and then the simulation would stop abruptly. That inspired me to do the same thing with SimCity for the Sun: you could play for a few minutes, then the simulator would switch to a cellular automata that would scramble your city, until you bought a license key to unlock it.
The GX was an 8-bit display, so it uses a 256 color palette, and drew shaded colors by halftone stippling. You can clearly see that in this video:
Aviator 1.5.1 running on SunOS 4.1.4 on a Sparcstation 20
https://www.youtube.com/watch?v=4RQscDJCy4c
Aviator had "aircraftcap" file in the format of (yech) a termcap file, that allowed you to configure your aircraft, and several fighter aircraft, small planes, and a 747 were included.
I heard a story about a fighter pilot who played Aviator, but he noticed something was wonky about how a plane he was familiar with flew. So he too a look at the aircraftcap file, and was able to find the problem, tweak it, and then it flew much more to his satisfaction!
You could dogfight over the network, and the sneaky thing to do was to hack your aircraftcap file to mount your guided missiles far behind you fighter jet, so when somebody was on your tail, you could shoot them in the ass!
Software in silicon: the methodology behind Sun's GX graphics accelerator architecture:
https://ieeexplore.ieee.org/document/63673
Sun GX Series Graphics Workstations:
https://old.hotchips.org/wp-content/uploads/hc_archives/hc01...
AVIATOR 1.5 FOR SUN NETWORKS OPENS UP GRAPHICS WORKSTATION GAMES MARKET (July 8 1991):
https://www.techmonitor.ai/hardware/aviator_15_for_sun_netwo...
SGI Users Group Forum: "Aviator" - a flight sim program for Sun GX - does anyone have this installed? Can anyone run it and get screenshots or video?
https://forums.sgi.sh/index.php?threads/aviator-a-flight-sim...
Vintage Computer Federation Forums: The "Aviator" flight sim for Sun GX hardware (e.g. SPARCstations)
https://forum.vcfed.org/index.php?threads/the-aviator-flight...
Reddit: A request to owners of Sun GX hardware such as SPARCstations: Can anyone post screenshots or footage from the late 80s/early 90s flight sim "Aviator"?
https://www.reddit.com/r/vintagecomputing/comments/y9zfhs/a_...
> At a time when PC memory maxed out at 640 megabytes,
Pretty sure the author meant write 640 kilobytes.
The sentence and paragraph which makes it clear that this was megabytes and not kilobytes
At a time when PC memory maxed out at 640 megabytes, the fact that the PCI bus could address 4 gigabytes meant that quite a few of its address bits were surplus. So we decided to increase the amount of data shipped in each bus cycle by using some of them as data. IIRC NV1 used 23 address bits, occupying 1/512th of the total space. 7 of the 23 selected one of the 128 virtual FIFOs, allowing 128 different processes to share access to the hardware. We figured 128 processes was plenty.
Okay but "640" is a completely fictitious number for installed RAM in any given PC.
PC memory was nearly always sold in powers of two. So you could have SIMMs in capacity of 1MiB, 2MiB, 4, 8, 16MiB. You could usually mix-and-match these memory modules, and some PCs had 2 slots, some had 4, some had a different number of slots.
So if you think about 4 slots that can hold some sort of maximum, we're thinking 64MiB is a very common maximum for a consumer PC, and that may be 2x32 or 4x16MiB. Lots of people ran up against that limit for sure.
640MiB is an absurd number if you think mathematically. How do you divide that up? If 4 SIMMs are installed, then their capacity is 160MiB each? No such hardware ever existed. IIRC, individual SIMMs were commonly maxed at 64MiB, and it was not physically possible to make a "monster memory module" larger than that.
Furthermore, while 64MiB requires 26 bits to address, 640MiB requires 30 address bits on the bus. If a hypothetical PC had 640MiB in use by the OS, then only 2 pins would be unused on the address bus! That is clearly at odds with their narrative that they were able to "borrow" several more!
This is clearly a typo and I would infer that the author meant to write "64 megabytes" and tacked on an extra zero, out of habit or hyperbole.
You are straight up wrong. The first computer I ever built was a Pentium 2, RivaTNT, and it had 640 MB RAM.
I can’t find the purchase receipts or specific board brand but it had four SDRAM slots, and I had it populated with 2x64 and 2x256.
Edit: Found it in some old files of mine:
I was wrong! Not four DIMM slots... three! One must have been 128 and the other two 256.
The article is a touch confusing, but I'm pretty sure I agree that they meant the 640 kilobyte limit of the OG PC architecture. The Pentium II dates from 1997, the NV1 to 1995, and the new PCI bus with its whopping 32-bits to 1992. 640MiB would have been a prodigious amount of memory at the time of launch.
I don't think any mathematical relationship between the address bus and either 640KiB or 640MiB was intended, it was just the anchor point for how huge 4GiB of addressing was viewed at the time.
The article then goes on to say that the NV1 used 23 bits of the address bus but adds in the next paragraph that 16 bits remained to use for data. That math isn't working out for me.
Actually, I'm really struggling to understand how this scheme would work at all. It strongly implies open addressing with no other MMIO devices to conflict with, but that's just not how I thought PCI worked. Maybe someone who knows more can explain it to me.
My reading was 640MiB was seen as some extraordinary upper bound that was unlikely to be breached in 1995 leaving lot of bits of the address bus for NV1 for quite some time. The 640 KiB seems definitely not a limit as even IBM PC/AT released in 1984! had an upper limit of 16MiB. So, as NV1's designer you could not assume 640KiB was some sort of upper bound on PC's of 1995 when designing the scheme. As to why 640MiB and not something else, I believe Windows 95 could address 2GB in theory but would start becoming unstable around 512MiB so maybe he chose 640MiB.
The whole thing a bit ironic since Bill Gates took great pains to say he never said 640KiB is all you need (or something like that). Given my example of the IBM PC/AT it definitely was not common understanding of upper limits in 1995 apocryphal or not.
To make this easy, there was no overlap between when 640kb was common on PCs and when nVidia existed as a company.
Yeah, I dunno. Besides being a lot for 1995, the address space stuff, if taken at face value, means you'd only have to knick 3 bits off before starting to eat into it. Shrug
By 1995, PCI had vanquished VL-Bus and a cheap entry PC would have had a PCI bus and a 486 CPU at 66 MHz or 100 MHz with 8 MB of RAM, while a decent PC would have had a Pentium with 16 MB of RAM.
The days of 640 kB memory were long forgotten.
Even in 1990, a decent PC would have had a 386 CPU with 4 MB of RAM, which would have been used as extended memory by the MS-DOS programs, while the cheapest computers would have been PC/XT clones with a NEC V20 CPU and with 2 MB of expanded memory, so still well beyond the 640 kB.
[dead]
Impressive build, probably sounded pretty wild upon startup with 4x ATA drives!
It did, especially as the drives got older and shrieky. Plus the floppy and optical drives spinning up. But... funny story, when I moved to Silicon Valley, SSDs didn’t exist yet, but compile times were IO-bound. So my company bought everyone Dell workstations with two 15k RPM SAS drives in RAID 0. When I’d kick off a build, the desk would start vibrating under my fingers.
It made everything feel real!
Later, the company switched new employees to 3x 10,000 RPM SATA drives. Not quite as grindy, but still loud.
Oh man, the Abit motherboards! That takes me back. How much did this cost and at what time? Presume very late 90s.
Looks like ‘97. Unfortunately I can’t find the receipts! If I had to guess, I think I’d say somewhere around $2000 in all. That computer lasted me a LONG time. When it was done being a desktop, it became a Linux server until I sold it around 2009. Sometimes I wish I’d kept it: the Pentium 2 was such an important CPU and I smile every time I see the one on display in the Computer History Museum.
Alright then! Humbly, I stand corrected about my poor speculation without research. It looks like 640MiB has been a perfectly achievable configuration, especially with 2x256+2x64 or such. That is, I must say, a huge amount of RAM. Like, way more than any video game ever specified in HW requirements. What use cases applied that you could use up 640MiB in that era, I wouldn't know!
I remain a bit mystified about why it would be a hard maximum, though. Did such motherboards prevent the user from installing 4x256MiB for a cool 1GiB of DRAM? Was the OS having trouble addressing or utilizing it all? 640MiB is not a mathematical sort of maximum I was familiar with from the late 1990s. 4GiB is obviously your upper limit, with a 32-bit address bus... and again, if 640MiB were installed, that's only 2 free bits on that bus.
So I'm still a little curious about this number being dropped in the article. More info would be enlightening! And thank you for speaking up to correct me! No wonder it was down-voted!
I did a bunch of media and software development back then so RAM helped a lot. Why 640? Not sure. My particular board could have gone up to 768. I did some googling and found some boards that maxed out at 1 GB.
That was a weird time in computing. Things were getting fast and big quickly (not that many years later, I built a dual-socket Xeon at 2.8 GHz, and before that my brother had a dual socket P3 at 700 MHz.) but all the expansion boards were so special-purpose. I remember going out of my way to pick a board with something like seven expansion slots.
But I think your question about why the author said 640 is fair! Maybe they had a machine like mine around then. Or maybe it’s something NVIDIA was designing around?
As another poster has said, around 2000/2001 there were many PC motherboards for which the maximum RAM capacity was 640 MB of SDRAM (though depending on the motherboard the capacity limit could have other values, e.g. 512 MB or 768 MB were other frequent values). Only around 2003, motherboards where you could install up to 4 GB of DDR memory became widespread.
I had a few such motherboards, with Pentium III "Coppermine" and with AMD Athlon "Thunderbird" (with 3 DIMM sockets, 2 accepting 256 MB modules and 1 accepting 128 MB modules).
While I had initially equipped them with less memory, i.e. either 128 MB or 256 MB, because of the high cost at that time, eventually I had maxed up the installed memory to 640 MB on the Athlon motherboards.
At that time, the memory controller was not included in the CPU, but it was contained in an external chip, the Northbridge. The MB chipsets could come from various vendors, including Intel and Via. Depending on the chipset, different maximum amounts of SDRAM could be supported. Sometimes there were also limits for the maximum amount of RAM caused by the L2 cache controller, due to the size of the tags required in cache lines. Even with the same chipset, the MB manufacturer could choose to route different numbers of address lines to the sockets, probably depending on whether the extra capacitive loading caused by a bigger memory module would have allowed it to function at the rated speed on that PCB.
64 MB would have been a typical maximum memory around 1998 (2 sockets for 32 MB SDRAM modules, on a MB with an AGP socket for the graphics card), so if the author would have had that year in mind, you would be right.
For a while, I had 36MB on a 486 mobo, with a combination of SIMMs and DIMMs. As others have mentioned, 640MB is a valid, if odd, configuration for RAM.
Not a single 486 motherboard supported DIMMs.
Haven't you ever heard the Bill Gates quote that he didn't actually say?
https://news.ycombinator.com/item?id=18120477
https://groups.google.com/g/alt.folklore.computers/c/mpjS-h4...
QUESTION: I read in a newspaper that in 1981 you said, ``640K of memory should be enough for anybody.'' What did you mean when you said this?
ANSWER: I've said some stupid things and some wrong things, but not that. No one involved in computers would ever say that a certain amount of memory is enough for all time.
The need for memory increases as computers get more potent and software gets more powerful. In fact, every couple of years the amount of memory address space needed to run whatever software is mainstream at the time just about doubles. This is well-known.
When IBM introduced its PC in 1981, many people attacked Microsoft for its role. These critics said that 8-bit computers, which had 64K of address space, would last forever. They said we were wastefully throwing out great 8-bit programming by moving the world toward 16-bit computers.
We at Microsoft disagreed. We knew that even 16-bit computers, which had 640K of available address space, would be adequate for only four or five years. (The IBM PC had 1 megabyte of logical address space. But 384K of this was assigned to special purposes, leaving 640K of memory available. That's where the now-infamous ``640K barrier'' came from.)
A few years later, Microsoft was a big fan of Intel's 386 microprocessor chip, which gave computers a 32-bit address space.
Modern operating systems can now take advantage of that seemingly vast potential memory. But even 32 bits of address space won't prove adequate as time goes on.
Meanwhile, I keep bumping into that silly quotation attributed to me that says 640K of memory is enough. There's never a citation; the quotation just floats like a rumor, repeated again and again.
It's hard to tell exactly what time frame the author is referencing there. For context, NV1 was released in '95, by which time it was not uncommon for a new PC to have 8-16 MB of memory (I had a 486 with 16 MB by '94). Especially if you planned to use it for gaming.
Maybe that is what they were thinking but anything designed to work with a PCI bus would have been introduced after PCs became capable of using more memory than that.
I still have a NV1 card.
Me too i also have a rendition verite card which i guess in some ways is the first real fully programmable consumer gpu as it has a risc processor.
I also have an even older NEC ISA card with a TI TMS34010 chip [1], that was a programmable CPU/GPU.
[1] https://en.wikipedia.org/wiki/TMS34010
Talk by Erik Lindholm (now retired), on Riva 128 and many of the other early Nvidia cards: https://ubc.ca.panopto.com/Panopto/Pages/Viewer.aspx?id=880a...
Thanks!
No mention of SGI.
AFAIK the SGI-Nvidia connection was after DSHR's time.
I thought the majority of SGI's contribution was braindead management decisions that caused the good engineers to leave and form nVidia.
[dead]
I stopped reading right here:
> Because Nvidia became one of the most valuable companies in the world, there are now two books explaining its rise and extolling the genius of Jensen Huang,
Yeah, he's a real genius. (Sarcasm). He is a marking guy, there is no genius behind this man.
The fact that Nvidia uses its market position cause harm to the industry by strong-arming partners to toe the line makes this company a problem, just like all the others. They operate like any other predatory corporation.
> He is a marking guy,
You would be wrong. He worked at AMD as a design engineer, and later went to LSI logic helping customers put out custom ASICs. One of his customers, a big customer, was Sun, helping with their SPARC processor and the GX graphics chips, and no doubt many others.
In 1989-1991 I did three ASICs at LSI Logic -- and Jensen was my liason there on the latter two. He was incredibly smart, hard working, technically knowledgeable, kind, patient, and generous with his time despite being very busy.
The marketing stuff came later (or maybe said better: it was latent and it came out later)
The Nvidia of the late 90s died once they realized people wanted compute on a video card.
Probably read the rest? I did not see Jensen's name on any of the patents that this key engineer discusses the detail and rationale of, and I feel that those names are listed fairly deliberately.